AWS Adds xAI’s Grok 4.3 to Bedrock for Long-Context Agent Workflows

The model brings a 1 million-token context window, configurable reasoning effort, image inputs and OpenAI-compatible access through Bedrock’s Mantle inference engine.

AWS has made xAI’s Grok 4.3 generally available on Amazon Bedrock, adding xAI as a model provider on the managed AI platform and giving Bedrock customers another option for long-document analysis, tool-using agents and multi-turn workflows.

Grok 4.3 accepts text and image inputs, returns text, and supports a 1 million-token context window. AWS positions those capabilities for enterprise tasks such as contract review, credit agreement analysis and financial document question answering, where a model may need to process large source materials before calling internal systems or external tools.

The model also offers per-request reasoning controls. Developers can select none, low, medium or high effort, allowing them to trade off latency against deeper reasoning within the same model. AWS suggests using no reasoning effort for classification tasks and higher settings for work such as contract or case-law analysis.

That control matters for production agent systems because a single workflow often contains both lightweight routing steps and expensive analytical steps. Teams can potentially avoid deploying separate models for each tier of work, though AWS did not provide pricing or latency figures that would show how those effort settings affect operating cost in practice.

Grok 4.3 runs on Mantle, Amazon Bedrock’s next-generation inference engine, rather than through the standard Amazon Bedrock Runtime API. AWS says developers can invoke the model with the OpenAI SDK or direct HTTPS requests through Mantle’s OpenAI-compatible APIs. For teams already using OpenAI-style client libraries and message formats, that could reduce integration work compared with adopting a provider-specific interface.

AWS emphasized the model’s tool calling and instruction-following capabilities, two features central to agents that need to retrieve records, invoke business functions or pass structured requests to other software. The model also supports structured output, stateful multi-turn conversations and image input, broadening its potential use in document workflows that combine scanned pages, tables, charts and follow-up questions.

xAI has promoted Grok 4.3 using several benchmark results. The company says it ranked first on Artificial Analysis’ Omniscience benchmark, with the lowest hallucination rate among the frontier models in that comparison. It also cites first-place results on the Artificial Analysis Tau2 Telecom benchmark for customer-support tool calling and on Vals AI’s Case Law and Corporate Finance benchmarks for document understanding.

Those results are useful signals for the workloads AWS is targeting, but they should not be treated as a complete deployment evaluation. AWS’s announcement attributes the benchmark claims to xAI and does not provide independent latency, reliability, price or throughput measurements for Bedrock usage. xAI’s claim that Grok 4.3 delivers two to 10 times more intelligence per dollar than other frontier models is likewise its own positioning, not a Bedrock cost comparison.

Parameter count information was not provided in AWS’s announcement.

For Bedrock customers, the practical change is provider choice paired with an unusually large context window and configurable reasoning settings in a managed AWS environment. The more consequential question will be whether Grok 4.3’s tool-use performance, long-context reliability and Mantle integration justify its inference cost against existing Bedrock options. AWS did not disclose regional availability, model pricing or enterprise adoption details in its launch announcement.